Industries such as pharma, manufacturing, logistics, retail, or any sector with a complex supply chain, often operate in a rapidly changing environment. A machine might suddenly break down, a delivery might be late, tasks may need rescheduling and so on. The person responsible needs to act on these events and make good decisions. However, this is often done manually by humans, leading to a number of issues.
Firstly, the decision process is not transparent as it happens in the mind of the decision-maker.
Secondly, because the decision depends on a lot of variables and humans might not be able to grasp these to the fullest, it’s often suboptimal.
Lastly, since conditions are constantly changing, intelligent decisions have to be made frequently and a lot of costly employee time is consumed.
Decision Intelligence (DI) is a solution to these problems. Gartner uses the following definition: “Decision intelligence is a practical discipline that advances decision making by explicitly understanding and engineering how decisions are made and how outcomes are evaluated, managed and improved via feedback.” This is a must-have for companies to enhance performance, improve operational efficiency and stay relevant in today's competitive landscape. According to Gartner, by 2026, 75% of global enterprises will apply decision intelligence, making augmented and automated decision making the next competitive differentiator.
In this article, we will go beyond a vague description of decision intelligence by showcasing how we have applied it for a global client in the manufacturing industry, CNH Industrial. The first paragraphs are about decision intelligence technology in general: what it is and what role AI plays in it. Next, we deep dive into the real-life use-case of setting safety stock for CNH Industrial. After a short description of the problem and the developed solution, we concretely demonstrate where decision intelligence adds value to an organization.
What is Decision Intelligence?
Let us revisit the definition by Gartner: “Decision intelligence is a practical discipline that advances decision making by explicitly understanding and engineering how decisions are made and how outcomes are evaluated, managed and improved via feedback.” So, by making an initial investment in understanding how you should arrive at a certain decision, you can engineer it such that it is automated. The added intelligence comes from the fact that you have this engineered process, but also from intelligent subcomponents in that process, be it forecasting using machine learning, AI agents that can reason or traditional planning optimization.
The term ‘decision’ is used here in a broad sense, it can range from daily operational decisions like ‘where should I drive to next?’ or ‘should I fix this broken machine now or later?’ to more strategic ones like ‘should I buy another truck?’ or ‘Can I handle demand surges of X% and if not where is the bottleneck?’. The idea is that decision intelligence can automate common decisions and assist in strategic ones.
Decision intelligence (DI) can be thought of as the new business intelligence (BI). BI provides static analytics and insights about past events, leaving humans to interpret these and drive decisions. DI on the other hand adds an intelligence layer, taking in the state the organization is in today and recommends optimal decisions for future scenarios. It is no longer a static report, real-time data might change the inputs or humans alter parameter settings, with new simulations or decisions as a result. In essence, we go from a paradigm of decisions made by humans guided by machines, to decisions made by machines guided by humans.
Broadly speaking there are three levels of DI, depending on the maturity of the engineered decision process or the trust you put in its recommendations:
Descriptive. A process or workflow is modeled as-is, allowing to simulate it under different conditions. The result offers insights to humans in various well-chosen scenarios.
Prescriptive. Decision making is fully automated, but there’s a human in the loop to verify its results and make the final call.
Autonomous. Decision making is fully automated and trusted. No human in the loop anymore.
Function of AI in decision intelligence
The engineered decision process often consists of multiple stages. Each stage can differ in complexity, from a simple fixed business rule to a more complex model. In the latter case, AI obviously plays an important role due to its advanced capabilities. For example, a first stage might be metadata extraction from images of contracts from the procurement department to retrieve price, supplier, logistics details, etc. This is made possible by AI using OCR & LLM models or vision language models (VLMs).
On top of modeling complex tasks, AI allows continuous learning. Since we live in a world that is constantly changing, models should catch up on these changing trends. By regularly retraining on the latest data, AI models can stay up-to-date, this in contrast to fixed business rules or one-time analyses that quickly become outdated.
AI domains that are particularly useful in DI:
Predictive AI: forecasting demand, predictive maintenance, forecasting risks …
Computer Vision: extraction of metadata, image classification, detection for inspection and quality control
Natural Language Processing: LLMs extracting information from documents, AI agents
Operations Research: machine/workforce/task planning, routing, any problem optimizing over a large number of combinations
Decision intelligence in practice: setting safety stock for a global manufacturing client
Problem setting
CNH Industrial manufactures complex agricultural machinery. In order to do so, the timely availability of all parts on the production line is of utmost importance. If a critical part would not be available and as a consequence the production line is interrupted, the associated cost is immense. To avoid this, safety stock acts as a buffer in inventory to absorb the uncertainty in the supply chain. The uncertainty arises from two sides. On the supply side, suppliers are not always able to deliver in time and on the demand side, fluctuations occur.
Setting appropriate levels of safety stock is challenging, since it must be done for tens of thousands of parts, for suppliers all around the globe having long lead times and based on a good quantification of the needed data: which suppliers are reliable and what is the demand variation.
For more details, refer to the case study.
Decision Intelligence Solution
The decision intelligence solution consists of three steps. First, the relevant historic delivery delays on the supply side and historic demand variation are extracted from the raw data. Since a new safety stock (or any decision really) affects a period of time in the future, we need to have an estimate of how likely delays and demand variation will be in the future. As such, in a second step, delays and demand variation are forecasted using AI. Now that we have an idea of supply and demand uncertainty for each part, we can set a safety stock. To do so, we start from a given budget and optimally spread it across all the parts in a final optimization step. This takes into account the forecasted uncertainty, price of the part and operational constraints (e.g. an engine is more important than a bolt).
Figure: Steps in determining safety stock
Benefits of decision intelligence
This section describes the most important aspect of DI: what value does it bring to the organization? The following benefits are applicable in general, not only in this use-case.
Transparency
Because the decision process is explicitly engineered, it is clear to everyone how a certain decision is reached. Before the added intelligence to the safety stock process, the person in charge would follow his/her gut feeling based on personal experience where most supplier issues are situated. Now, the different steps to arrive at a certain safety stock are known and supported and made visible by data along the way: what were the delays of the supplier, what is the demand variation, what are their forecasts… This removes the bias of individuals.
Also, the definitions of the concepts, delays and demand variability, become transparent. Everyone has a notion of their meaning, but people implicitly use their own definition. Aligning and defining these concepts within the organization creates clarity and transparency.
Automation at scale
Automating the decision process makes it possible to run at scale at a limited cost. For large complex supply chains, scale is a must, as in this case, we are handling tens of thousands of parts. Scaling up is also easier. Roughly speaking, it’s just the computer running for longer instead of the employee spending more time. Whereas before, reviewing the existing safety stock would be a manual and repetitive task, now it boils down to the press of a button. Freed up time can be used to verify or gain insights in its recommendations, or simply trust it and spend elsewhere.
Higher performance
Decisions often depend on a lot of factors. It’s impossible for humans to take the sheer quantity of these into account using their up-to-date operational values. For example, the optimization step distributes a given budget over tens of thousands of parts. The amount of possibilities, e.g. X units for part M, Y units for part N etc, is intractable for a human to optimize.
Purely the fact that the decision process is mapped out, already leads to better decisions. When supply chain planners would set the safety stock, they would never follow the same process, leading to suboptimal performance. Also the subcomponents in itself, AI forecasting and optimization, are more performant than a human could accomplish.
Due to these advantages, the solution achieves at least a 10% reduction in express deliveries and 50% reduction in parts missing on the production line, for one of the pilot plants.
Faster decisions
Thanks to automation instead of manual work, routine decisions can be made in a matter of minutes instead of hours. The same is true for determining safety stock. Because of this low effort, it has now become possible to rerun the workflow after certain events or changes have occurred in the supply chain. This means the state of your organization can adapt more quickly to reality.
Continuous learning
The underlying AI models are retrained regularly on the latest data. As the environment and thus the supply chain changes in behavior, the underlying models adapt to follow the latest trends. In the case of safety stock, this means the forecasting models adapt. In volatile supply chain times, it will forecast higher chances of delays and vice versa in stable times.
Conclusion
Decision Intelligence (DI) engineers and thus automates the decision process, with multiple benefits as a consequence and ultimately leading to increased operational efficiency and better strategic insights. Rather than relying on gut feeling, organizations can now use data and AI to consistently make better decisions.
In this article, we’ve made the concept of DI tangible through a practical use-case in manufacturing. By tackling the challenge of setting safety stock for CNH Industrial, we’ve shown how DI can be applied in the real world. Not as a theoretical concept, but as an operational tool that delivers concrete results.
The benefits are clear. Decision processes become transparent and aligned across teams. Automation allows operations to scale at minimal cost. Performance improves through the added intelligence, in which AI plays a big role. Routine decisions can be made faster, freeing up time. And finally, the continuous learning capabilities of AI ensure decisions remain relevant in a dynamic environment.
We, at Superlinear, have expertise and experience in making your complex value chains smarter, see this article on how to implement a roadmap. Interested in identifying a first use-case? Contact us. We’re happy to help.
FAQs about decision intelligence
How to get started with decision intelligence?
Start by identifying where decision-making is reliant on gut feeling and time intensive, while prioritizing high impact use-cases. Make sure data is being collected. Next, engineer the decision process and automate it. Monitor the generated decisions and refine where needed. For a more detailed guide, refer to this roadmap.
Does decision intelligence replace humans?
In short: not at all. First of all, humans are required to engineer the decision process. Secondly, decision intelligence is meant to relieve humans of repetitive and time-consuming tasks. Instead, they can focus on insights and improving the decision-making process.
Where does decision intelligence thrive?
Decision intelligence thrives in complex environments that are constantly changing and where data keeps track of information flows.
What are the components of a decision intelligence platform?
Roughly speaking, we can identify 4 main components:
A trusted, qualitative data layer acts as the foundation. Without it, people will be hesitant to trust the generated decisions. Not because the intelligence is insufficient, but simply because the data fed into it does not reflect reality.
A data transformation layer extracting higher level information, such as metadata from documents, concepts specific to the organization etc.
The intelligence layer. It combines the transformed data in a smart way, using forecasting, traditional optimization and AI agents that can reason. The generated decisions can in turn be used as input to other decision intelligence modules, creating a modular and complex framework.
Interactive visualizations. This allows humans to gain insight, guide the intelligence layer where needed and monitor its decisions.
What is the difference between decision intelligence and business intelligence?
Business intelligence provides static analytics and insights about past events, leaving humans to interpret these and drive decisions. Decision intelligence on the other hand adds an intelligence layer, taking in the state the organization is in today and recommends optimal decisions for future scenarios. It is no longer a static report.
